• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学 ›› 2024, Vol. 46 ›› Issue (09): 1529-1538.

• 高性能计算 • 上一篇    下一篇

面向结构矩阵的可扩展并行矩阵乘算法框架

李胜国1,廖霞2,于恒彪1,黄春1,姜浩1,逯喜燕1,王华林3,成礼智3   

  1. (1.国防科技大学计算机学院,湖南 长沙 410073;2.清华大学计算机科学与技术系,北京 100084;
    3.国防科技大学理学院,湖南  长沙  410073)

  • 收稿日期:2023-01-04 修回日期:2023-07-04 接受日期:2024-09-25 出版日期:2024-09-25 发布日期:2024-09-19
  • 基金资助:
    国家重点研发计划(2021YFB0300101)

A scalable parallel structured matrix multiplication algorithm framework

LI Sheng-guo1,LIAO Xia2,YU Heng-biao1,HUANG Chun1,JIANG Hao1,LU Xi-yan1,WANG Hua-lin3,CHENG Li-zhi3   

  1. (1.College of Computer Science and Technology,National University of Defense Technology,Changsha 410073;
    2.Department of Computer Science and Technology,Tsinghua University,Beijing 100084;
    3.College of Science and Technology,National University of Defense Technology,Changsha 410073,China)
  • Received:2023-01-04 Revised:2023-07-04 Accepted:2024-09-25 Online:2024-09-25 Published:2024-09-19

摘要: 摘要:结构矩阵在科学计算和工程应用中具有重要作用,例如Cauchy、Toeplitz、Vandermonde和Hankel矩阵等。虽然这些矩阵都是稠密的,但只需要O(n)个参数(生成元)就可以表示,其中n为矩阵的维数。提出了面向结构矩阵的可扩展并行矩阵乘算法框架,利用矩阵生成元显式地构造各进程的局部矩阵块,从而减少通信开销;同时利用矩阵块的数值低秩性,进一步降低计算开销。因此,该算法框架可同时降低计算量和通信量,适用于Cannon、Fox和PUMMA等矩阵乘算法。在天河2巨型机上进行了大量的数值测试,测试结果表明,该算法可获得相对ScaLAPACK中的PDGEMM函数的8.96倍加速。

关键词: 结构矩阵, 矩阵乘法, FFT, Cauchy矩阵, Toeplitz矩阵, 分布式并行

Abstract: Structured matrices play an important role in scientific computing and engineering applications, such as Cauchy, Toeplitz, Vandermonde, and Hankel matrices. Although these matrices are dense, they can be expressed with only O(n)  parameters (generators), where n is the dimension of the matrix. The core idea of the algorithm in this paper is to use matrix generators to explicitly construct local matrix blocks of each process, thereby reducing communication overhead. Additionally, by leveraging the numerical low-rank property of these matrix blocks. This paper further minimize computational overhead. Consequently, the proposed parallel structured matrix multiplication algorithm framework can simultaneously reduce both computational and communication costs, making it suitable for matrix multiplication algorithms like Cannon, Fox, and PUMMA. Extensive numerical tests were conducted on the Tianhe-2 supercomputer, and the results demonstrate that the proposed algorithm achieves an 8.96× speedup compared to the PDGEMM function in ScaLAPACK. 

Key words: structured matrix, matrix multiplication, FFT, Cauchy matrix, Toeplitz matrix, distributed parallel